VoiceArticulation

Twister Marathon

The scored articulation event. Twisters run back-to-back, three repetitions each, with difficulty ramping through the session — and unlike the other two drills, the app knows exactly what you were supposed to say, so it can grade how well you said it.

Structure
3 reps per twister, ramping
Advance
Recognition or karaoke pacing
Score
0.25 pace + 0.75 diction
Stars
Accuracy + pace gate
The method

Why it works

Reps are what turn a twister from a puzzle into articulation training. Three passes at the same phrase let you find the failure, fix it, and prove the fix — and because the ramp mixes difficulty levels, the session gets harder as your mouth gets warmer.

  • A phrase you cannot fake

    The reference text is known, so the score comes from word-level alignment against it, not from an impression of how you sounded.

  • Recognition-driven advance

    The card moves on when recognition both matches closely enough and has reached the end of the phrase. A ratio alone is not a completion signal — hitting two words of a three-word twister already scores 0.67 and used to switch the card early.

  • Difficulty-aware pace

    Twisters are meant to be spoken faster than conversation, so the pace target shifts up with difficulty instead of grading you against a chat.

How a session runs

Step by step

  1. 1

    Twisters are dealt in a ramp, three repetitions of each.

  2. 2

    Say the phrase; the card advances when what was recognised reaches the end of it and matches closely enough.

  3. 3

    Karaoke pacing is available if you would rather the timeline advance the reps for you.

  4. 4

    At the end you get session scores and a 0–3 star rating.

Evaluation

How it's scored

Diction is measured by aligning what was recognised against the twister you were given, word by word. Pace is a Gaussian around a target that rises with the difficulty you actually faced. Overall weighs them 25/75 — clarity is the exercise, speed is the garnish.

Deterministic · twister-marathon-performance

Diction

75%

Weighted word alignment against the reference phrase: a correct word scores 1.0, a partial match 0.5, a missing or substituted word 0. The result is your recognised accuracy as a percentage.

Pace

25%

A Gaussian on words per minute around a language target (135 EN / 115 RU) boosted by the difficulty mix the session actually contained: +30 for High, +15 for Medium, +0 for Low, averaged over the twisters that ran.

The formula

Transcribed from the scorer that ships in the app. Every constant is pinned to this methodology version, so the same take always produces the same number.

Diction
diction = (correct×1.0 + partial×0.5) / expectedWords × 100
Pace
target = languageTarget + difficultyBoost      // 135 EN / 115 RU
sigma  = 35 (EN) | 30 (RU)
pace   = 100 × exp(−(wpm − target)² / (2σ²))
Overall
overall = 0.25 × pace + 0.75 × diction
Stars
3 ★  accuracy ≥ 95%  and  wpm ≥ threeStarWpm
2 ★  accuracy ≥ 85%
1 ★  accuracy ≥ 70%
0 ★  otherwise

threeStarWpm = 150 (High) | 120 (Medium) | 90 (Low),
               averaged across the levels the ramp contained

Details worth knowing

  • The methodology spec's small cleanliness weight is folded into diction: tongue twisters carry no meaningful filler signal.
  • Pace is always reported — it is an honest audio measurement — even when diction is not.

When no score is given

A weak guess is worse than no number, so the overall is withheld rather than estimated. What was measured stays visible either way.

  • A fully karaoke-paced session gets no diction and no overall: the reps were advanced by the pacing timeline rather than by recognition, so there is no alignment evidence to grade. The number is withheld rather than fabricated from the fixed karaoke credit.

Reading the number

85–100Excellent
70–84Good
50–69Average
Below 50Needs practice
The coach review

What the AI adds

The coach review treats a marathon as a pronunciation exam:

  • How you handled the specific consonant and vowel clusters of the twisters you drew.

  • Per-phoneme-group articulation scores backed by words from the take.

  • Stumbles and restarts *within* a twister — pauses *between* them are explicitly not counted against you.

Try Twister Marathon Yourself

Download Eloqo, open this exercise, and speak. That's the whole first session.

In closed testingGoogle PlayComing soon to theApp Store

Eloqo is in closed testing while we finish the public release.See the app in action →